Drainage basin real-time flood control dispatching risk dynamic early warning and emergency scheme making method

Abstract

The invention discloses a drainage basin real-time flood control scheduling risk dynamic early warning and emergency scheme making method, which comprises the steps of obtaining a real-time total element information tensor and an engineering flood control design parameter set, and calculating to obtain a real-time hierarchical risk entropy time sequence front section based on safety margin distribution, and inputting real-time data and pre-constructed entropy flux Markov graph structure data into the trained entropy perception double-flow prediction model for reasoning to obtain a future risk prediction set. And a dynamic risk early warning result set is generated in combination with a preset entropy level threshold and a future failure chain probability, and an emergency scheduling scheme set is generated on the basis of meeting safety margin constraints and entropy reduction targets. According to the invention, dynamic early warning and decision making of strong physical-risk coupling are realized.

Description

Technical Field

[0001] This invention belongs to the field of smart water conservancy and flood control safety technology, especially the method for real-time flood control scheduling risk dynamic early warning and emergency plan formulation in watersheds. Background Technology

[0002] Faced with the frequent occurrence of extreme rainfall events caused by global climate change, traditional scheduling models based on experience and static rules are becoming increasingly inadequate. Therefore, researching dynamic risk early warning and emergency decision-making methods for real-time flood control scheduling in river basins, and realizing the transformation from passive response to proactive prediction, has significant theoretical and practical implications for improving the flood control resilience and emergency response capabilities of river basins.

[0003] Currently, research on river basin flood control scheduling mainly focuses on two aspects: First, flood evolution forecasting based on hydrodynamic models, using high-precision numerical simulations (such as HEC-RAS and MIKE models) to predict river levels, flows, and inundation extent in future periods. Second, risk indicator prediction based on data-driven models, using statistical methods or conventional time-series networks (such as LSTM) to fit historical monitoring data and predict physical quantities such as water levels and flows at single or a few key sections. Meanwhile, in terms of risk assessment, existing methods mostly employ multi-index fuzzy evaluation, failure probability statistics, or scenario analysis to conduct post-assessment of the risk consequences of scheduling schemes.

[0004] Existing solutions have significant shortcomings in the physical connotation of risk measurement, the spatiotemporal propagation mechanism of risk, and the physical-risk coupling of prediction models. These problems lead to delayed early warnings, a black box effect on risk evolution, and difficulty in supporting precise emergency dispatch. First, risk measurement is disconnected from the physical state. Traditional risk is mostly defined based on historical statistical failure probabilities, which have limited characterization capabilities and are difficult to correlate with the core physical state of the project—the real-time safety margin (i.e., the difference between design bearing capacity and real-time load). This results in risk assessment results failing to intuitively reflect the project's real-time safety reserves. Second, there is a lack of quantitative models for the spatiotemporal coupling and propagation of risk. Existing methods treat each engineering node as an independent risk source, ignoring the fact that risk flows, accumulates, and dissipates like entropy in the watershed topology (e.g., from upstream reservoirs to downstream dikes). This lack of a spatiotemporal propagation mechanism makes it impossible to accurately predict cascading failure chains where instability in one area leads to widespread emergency. Finally, flood evolution prediction is disconnected from the evolution of risk states. Data-driven models (such as LSTMs) are trained only to fit physical curves such as water level and flow rate, and their internal states do not contain any risk information. This makes the models unable to perceive the nonlinear impact of physical processes (such as flood peaks) on risk states (such as the increase in safety margin entropy), and also unable to use risk constraints (such as entropy conservation) to correct their physical predictions, resulting in predictions that are known but not understood. Summary of the Invention

[0005] The purpose of this invention is to provide a method for real-time flood control scheduling risk dynamic early warning and emergency plan formulation in a river basin, in order to solve one of the problems mentioned above in the existing technology.

[0006] Technical solution: A method for real-time flood control scheduling risk dynamic early warning and emergency response plan formulation in a river basin, including:

[0007] Obtain real-time full-element information tensor and engineering flood control design parameter set; calculate the real-time safety margin state set and the first segment of the real-time hierarchical risk entropy time series based on this;

[0008] The real-time full-element information tensor, the real-time safety margin state set, and the pre-constructed entropy flux Markov graph structure data are input into the trained entropy-aware dual-flow prediction model to perform inference and obtain a future risk prediction set that includes the future hierarchical risk entropy prediction result set, the future entropy flux prediction result set, and the future failure chain probability distribution set.

[0009] By combining the real-time hierarchical risk entropy time series, the future risk prediction set, and the preset entropy level threshold, a dynamic risk warning result set is generated.

[0010] Based on the dynamic risk early warning result set, the future risk prediction set, and the engineering flood control design parameter set, and in accordance with the safety margin constraints and entropy reduction objectives, an emergency dispatch scheme set is generated.

[0011] Beneficial effects include: introducing physical safety margin as an anchor point for risk quantification; constructing a three-level risk entropy system from single project to subsystem to system by calculating the relative margin between engineering design capability and real-time load, thus giving risk assessment a solid physical basis; constructing an entropy flux Markov graph model to transform isolated node risks into a spatiotemporally coupled dynamic propagation process, realizing quantitative prediction of the evolution path of cascading failure risks; and designing an entropy-aware dual-flow prediction model, which achieves dynamic early warning and decision-making with strong physical-risk coupling through parallel processing of physical process flow and risk structure flow. Attached Figure Description

[0012] Figure 1 This is the overall flowchart of this application.

[0013] Figure 2 This is a flowchart of the process for calculating the real-time safety margin state set and the real-time hierarchical risk entropy time series for this application.

[0014] Figure 3 This is a flowchart for calculating the time series set of safety margins for a single project in this application.

[0015] Figure 4 This is a flowchart for obtaining the single-project safety margin interval label set for this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the following description will be provided in conjunction with the appendix. Figure 1 To be continued Figure 4 The present invention will be further described in detail below, along with specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0017] Example 1 provides a basic architecture and data preparation process for implementing the method described in this invention, which provides a unified and reliable data foundation for the calculation of risk entropy, model construction, and real-time extrapolation in all subsequent examples.

[0018] In this embodiment, the acquisition and full-element information tensor of multi-source data for the watershed is first constructed. Specifically, this process reads the basic watershed multi-source monitoring raw dataset from data sources such as hydrological monitoring stations, rain gauges, remote sensing platforms, and engineering automation systems within the watershed. This raw dataset is massive and heterogeneous, and for example, it includes at least: real-time and historical rainfall sequences, river flow sequences, water level sequences, reservoir capacity sequences, gate opening sequences, generator output sequences, and related basic meteorological elements.

[0019] After obtaining the raw dataset, rigorous quality control is required. Quality control operations may include: identifying missing values ​​in the data sequence, removing outliers exceeding reasonable thresholds, diagnosing sensor drift over time, and using appropriate interpolation algorithms to imputate and reconstruct missing data. After these operations, a quality-controlled monitoring dataset with a controlled error range is obtained. Interpolation algorithms can include linear interpolation, kriging interpolation, or machine learning models.

[0020] Based on this, the spatiotemporal reference is unified and assimilated. Since data from different sources may use different coordinate systems and time standards, they must be unified. First, based on a unified national or regional coordinate system (e.g., CGCS2000), the geographic coordinates of each data source are transformed to a unified spatial reference, resulting in a spatially aligned observation dataset. Then, according to the time zone information and observation period of the monitoring equipment, the timestamps from different sources are synchronized to a unified standard time axis (e.g., UTC+8), constructing a time-aligned observation dataset. To further improve data quality, data assimilation methods can be used, such as ensemble Kalman filtering (EnKF), to jointly estimate the time-aligned observation dataset with the output of existing watershed hydrological models (e.g., SWAT models or HEC-RAS models), thereby effectively correcting observation noise and model bias, forming an assimilated watershed state dataset under a unified spatiotemporal reference.

[0021] This study extracts engineering and exposure information and establishes node attributes to assess the disaster-bearing capacity and socio-economic vulnerability of engineering projects within the watershed. Specifically, it reads the assimilated watershed state dataset and, in conjunction with reviewing the design specifications, as-built drawings, and historical operational records of various engineering projects (such as reservoirs, dikes, dams, and pumping stations), extracts key structural parameters, design flood control standards (such as the design peak flow or design allowable water level corresponding to a 100-year flood event), maximum flood discharge capacity, and historical dispatch rules to form a set of engineering flood control design parameters describing the disaster-bearing capacity of the projects. Simultaneously, based on the latest land use type data, spatialized population density distribution maps, and GDP statistics, it constructs a set of exposure and vulnerability parameters sensitive to flood consequences. Finally, it associates the engineering flood control design parameter set and the exposure and vulnerability parameter set with their corresponding engineering nodes and spatial units in the watershed topology, respectively, forming an engineering and exposure attribute table describing the static attributes of each engineering node and watershed unit.

[0022] A comprehensive information tensor is constructed. The assimilated watershed state dataset, the engineering and exposure attribute tables, and the pre-defined watershed spatial grid partitioning scheme are read. The dynamic features corresponding to each time step are mapped to the corresponding watershed grid points and engineering nodes. Simultaneously, the engineering flood control design parameter set and the exposure and vulnerability parameter set are encoded as static features and attached to the corresponding nodes or grid points. The dynamic features include rainfall, inflow, water level, reservoir capacity, gate opening, and generator output. In this way, all information is arranged and combined into a multi-dimensional array with time, space, and feature dimensions—the comprehensive information tensor—to centrally express the hydrological, engineering, and exposure states of the watershed at each time step. This serves as the unified input data for safety margin calculation, hierarchical risk entropy construction, and predictive modeling in subsequent embodiments.

[0023] Example 2: This embodiment provides an overall process for a method of dynamic early warning of real-time flood control scheduling risks and the formulation of emergency plans for river basins.

[0024] Step 201: Obtain the real-time full-element information tensor and the engineering flood control design parameter set; based on this, calculate the real-time safety margin state set and the first segment of the real-time hierarchical risk entropy time series.

[0025] Specifically, the acquisition and construction methods of the real-time full-element information tensor and the engineering flood control design parameter set have been described in detail in Example 1. During the real-time operation phase, the system continuously receives the latest monitoring data and updates the real-time full-element information tensor in a rolling manner according to the data processing flow described in Example 1.

[0026] Based on this, the hierarchical risk entropy construction method described in subsequent embodiments is invoked.

[0027] Specifically, based on the real-time full-element information tensor and the engineering flood control design parameter set, a single-project safety margin time series reflecting the relative safety level of each project is calculated; the risk entropy of a single project is defined and calculated based on the probability distribution of the safety margin; and the risk entropy of subsystems and the system are constructed level by level, combining the topological coupling relationship and exposed vulnerability between projects. The calculated safety margin-related data (e.g., safety margin values ​​and their discrete interval labels) are collected into the real-time safety margin state set, and the risk entropy sequences of each level (single project, subsystem, system) are collected into the first segment of the real-time hierarchical risk entropy time series.

[0028] Step 202: Input the real-time full-element information tensor, the real-time safety margin state set, and the pre-constructed entropy flux Markov graph structure data into the trained entropy-aware dual-flow prediction model to perform inference and obtain a future risk prediction set containing a future hierarchical risk entropy prediction result set, a future entropy flux prediction result set, and a future failure chain probability distribution set.

[0029] Specifically, the entropy flux Markov graph structure data is pre-constructed based on historical data before model training. It describes the pattern of risk entropy transmission and the probability of state transitions in the watershed topology network. The entropy-aware dual-flow prediction model that has been trained is also pre-trained.

[0030] During the real-time operation phase, the real-time full-element information tensor (reflecting the physical process) and the real-time safety margin state set (reflecting the risk state) within the current moment and the near historical window are used as inputs, while the entropy flux Markov graph structure data is loaded simultaneously. After receiving these inputs, the entropy-aware dual-stream prediction model performs a forward inference and outputs predictions for several future time steps (e.g., the next 1 hour, 3 hours, and 6 hours). These prediction results are organized into the future risk prediction set, which includes predictions of future hierarchical risk entropy (single project, subsystem, system), predictions of entropy flux on the critical path, and predictions of the probability distribution of failure of critical project combinations.

[0031] Step 203: Combine the real-time hierarchical risk entropy time series, the future risk prediction set, and the preset entropy level threshold to generate a dynamic risk warning result set.

[0032] Specifically, a set of future hierarchical risk entropy prediction results and a set of future failure chain probability distributions are extracted from the future risk prediction set. Then, these predicted risk entropy values ​​are compared with pre-set entropy level thresholds. These entropy level thresholds are determined based on historical extreme events and the flood control objectives of the basin, for example, dividing risk entropy into four levels: reliable zone, alert zone, crisis zone, and critical zone. By comparison, it can be determined which level of risk entropy each time point and each engineering node will be at in the future. Combining the failure risk of key projects (such as upstream reservoirs) indicated by the future failure chain probability distribution set, the warning level is corrected or confirmed. A dynamic risk warning result set containing a time dimension (each future time point) and a spatial dimension (each engineering node and subsystem) is generated to issue clear risk alerts to dispatchers.

[0033] Step 204: Based on the dynamic risk early warning result set, the future risk prediction set, and the engineering flood control design parameter set, and in accordance with the safety margin constraints and entropy reduction objectives, generate an emergency dispatch scheme set.

[0034] Specifically, when the dynamic risk warning result set indicates an increased risk level (e.g., entering a alert zone or crisis zone), the system automatically triggers the generation of an emergency dispatch plan. The system will first deduce a series of candidate emergency dispatch operations that meet the constraints of the engineering flood control design parameter set (e.g., the reservoir must not exceed the flood limit water level, the gate opening speed is limited, etc.), such as combinations of operations such as pre-release, peak shaving, flood diversion, or staggered peak dispatch.

[0035] The system evaluates the effectiveness of these candidate solutions. The evaluation is based on the following criteria:

[0036] Safety margin constraints mean that scheduling operations cannot cause the safety margin of other projects to fall below a set lower limit;

[0037] The entropy reduction objective is that scheduling operations should effectively reduce the predicted future risk entropy level.

[0038] The entropy flux prediction results from the future risk prediction set can be used to simulate the changing trends of future entropy flux and hierarchical risk entropy after executing different candidate schemes. In this way, the optimal scheduling schemes that can maximize entropy reduction while ensuring the safety of all projects can be selected, forming the final emergency scheduling scheme set, which can be discussed, confirmed, and issued by the scheduling personnel.

[0039] Example 3: This example describes in detail the specific solution process of the single project safety margin time series set described in step 201 of Example 2. This process is the physical basis for all subsequent risk entropy calculations.

[0040] Step 301: For each engineering node, extract its engineering design flood control capacity from the engineering flood control design parameters set.

[0041] Specifically, this step corresponds to the process of constructing the engineering flood control design parameter set in Embodiment 1. In this embodiment, for each monitored and scheduled engineering node (e.g., a reservoir dam, a section of dike, a floodgate), it is necessary to query and extract its key design flood control capacity parameters from the engineering flood control design parameter set.

[0042] The engineering design flood control capacity C_design_e is a comprehensive index used to characterize the ultimate bearing capacity of an engineering project. Its specific meaning varies depending on the type of project. For example, for reservoir projects, C_design_e can be its design flood control level (e.g., 80.0 meters) or the reservoir capacity corresponding to the check flood level; for dike projects, C_design_e can be its design allowable water level (e.g., 75.0 meters); for spillway projects, C_design_e can be its design maximum flood discharge capacity (e.g., 8000 cubic meters per second). These extracted capacity parameter values ​​are associated with the unique identifiers of the engineering nodes to form an engineering flood control capacity parameter table for subsequent calculations.

[0043] Step 302: Extract the real-time actual flood load of the corresponding engineering node from the real-time full-element information tensor.

[0044] Specifically, the real-time actual flood load L_load_e(t) is a dynamic quantity that changes with time t, used to characterize the actual load borne by the project at the current moment. This load data needs to be extracted from the real-time full-element information tensor constructed in Example 1, based on the identifier of the project node and the current time step t.

[0045] It should be noted that the real-time actual flood load L_load_e(t) is not a single observation value, but an equivalent load defined according to the engineering failure mechanism. For example:

[0046] For reservoir-type projects, L_load_e(t) may not only be the real-time water level, but also the superposition value of the real-time water level and the additional water level caused by wind and waves.

[0047] For levee projects, L_load_e(t) may be the equivalent external load calculated after comprehensively considering the real-time water level, river flow velocity and wave impact pressure.

[0048] For dam and gate projects, L_load_e(t) may be the total stress on the gate leaf and foundation calculated based on the difference in water levels between upstream and downstream and the degree of gate opening.

[0049] This refined definition of L_load_e(t) overcomes the limitations of traditional methods that only use a single water level or flow rate index, making the calculation of the safety margin closer to the actual stress state of the project.

[0050] Step 303: Based on the project's designed flood control capacity and the real-time actual flood load, calculate the safety margin value and aggregate the safety margin values into the single-project safety margin time series set.

[0051] Specifically, after obtaining the project's designed flood control capacity C_design_e and the real-time actual flood load L_load_e(t) of project e at time step t, the current safety margin value M_safe_e(t) can be calculated through a safety margin definition formula.

[0052] In this embodiment, a calculation formula in the following form is preferably adopted:

[0053] M_safe_e(t)=(C_design_e-L_load_e(t)) / C_design_e;

[0054] The safety margin M_safe_e(t) represents the relative margin of the current load L_load_e(t) from the design capacity C_design_e.

[0055] When L_load_e(t)<C_design_e, M_safe_e(t)>0, indicating that the project is in a safe state. The larger the value, the safer it is.

[0056] When L_load_e(t)=C_design_e, M_safe_e(t)=0, indicating that the project is in the design critical state.

[0057] When L_load_e(t)>C_design_e, M_safe_e(t)<0, indicating that the project is in an overloaded state, that is, the margin is negative and it faces the risk of failure.

[0058] As an example, assume that the designed flood control level C_design_e of a certain reservoir is 80.0 meters. At time step t1, its equivalent flood load (real-time water level + wind wave addition) L_load_e(t1) is 78.5 meters. Then the safety margin at this moment is:

[0059] M_safe_e(t1)=(80.0-78.5) / 80.0=1.5 / 80.0=0.01875;

[0060] At time step t2, the flood is rising and the equivalent flood load L_load_e(t2) reaches 80.2 meters. Then the safety margin at this moment is: M_safe_e(t2) = (80.0 - 80.2) / 80.0 = -0.2 / 80.0 = -0.0025;

[0061] By repeating the above calculations for each engineering node e at all real-time time steps t, a series of safety margin values varying with time can be obtained. Organizing these values in engineering and time order forms the single-engineering safety margin time series set. This time series set will be used as the direct input for subsequent calculation of the single-engineering risk entropy.

[0062] In some alternative embodiments, the calculated safety margin time series can also be corrected. For example, check whether the safety margin value is within a reasonable range, such as whether it is between -0.1 and 0.5. If the value far exceeds the expected threshold, it may be caused by sensor anomalies. In this case, it can be marked as a suspected outlier and corrected by interpolation of adjacent time steps or truncation to a preset boundary value to enhance the robustness of subsequent entropy calculations.

[0063] Example 4: This example describes how to generate a single-engineering risk entropy time series set and a single-engineering safety margin interval label set based on the single-engineering safety margin time series set calculated in Example 3 by estimating the probability distribution of the safety margin interval.

[0064] Step 401, discretize the safety margin values in the single-engineering safety margin time series set according to the preset safety margin interval division rules. Map the continuous safety margin value M_safe_e(t) to a finite discrete interval with clear physical meaning.

[0065] The safety margin interval division rule set is preset and used to define the risk level.

[0066] Exemplarily, this rule set can divide the value range of the safety margin M_safe_e(t) into four intervals k = 1, 2, 3, 4, for example:

[0067] Interval 1, negative margin interval: M_safe_e(t) <= 0. The corresponding project is overloaded and in a high-risk state.

[0068] Interval 2, critical margin interval: 0 < M_safe_e(t) <= 0.02. The corresponding project has a very low margin and is close to the critical state.

[0069] Interval 3, low margin interval: 0.02 < M_safe_e(t) <= 0.05. The corresponding project has insufficient margin and is in a state of alert.

[0070] Interval 4, a margin-sufficient interval: M_safe_e(t)>0.05. This corresponds to engineering safety and a reliable state.

[0071] These thresholds (such as 0.02, 0.05) were determined based on the requirements for flood control and safety management in the basin and historical experience.

[0072] In some alternative implementations, the interval division rules do not need to be fixed. For example, different thresholds can be set according to different project categories (such as reservoirs and dikes). Alternatively, a data-driven approach can be adopted, such as using statistical methods (such as percentile methods) or clustering algorithms in machine learning (such as K-Means algorithm) based on historical safety margin time series sets to automatically determine the optimal interval boundaries, so that the interval division can better reflect the actual distribution characteristics of the data.

[0073] Step 402: At each time step, the conditional probability of the safety margin value falling into each safety margin interval is calculated to obtain the probability distribution of the safety margin interval.

[0074] Specifically, the purpose of this step is to estimate the uncertainty of the safety margin M_safe_e(t) near the current time step t. Directly using the instantaneous value of M_safe_e(t) (such as 0.01875) results in information loss because it does not reflect whether the value is stable at 0.01875 or fluctuates wildly between 0 and 0.04.

[0075] To estimate this uncertainty, this embodiment employs a sliding time window method. Specifically, a sliding time window is constructed for each project node e, for example, setting the window length to W (e.g., W=60, representing the past 60 time steps, such as 60 minutes). At each current time step t, W safety margin samples for the project are extracted within the [t-W+1,t] time window.

[0076] Based on the interval division rules defined in step 401, the frequencies N_k of these W samples falling into each interval k=1,2,3,4 are counted. The conditional probability p_e_k(t) within this time window is obtained by frequency normalization:

[0077] p_e_k(t) = N_k / W;

[0078] For example, at time step t, out of 60 samples (W=60) for a certain project e, 6 fall into interval 1, 30 into interval 2, 24 into interval 3, and 0 into interval 4. Then, its safety margin interval probability distribution at time t is:

[0079] p_e_k(t)=[p_e_1(t),p_e_2(t),p_e_3(t),p_e_4(t)]=[6 / 60,30 / 60,24 / 60,0 / 60]=[0.1,0.5,0.4,0.0];

[0080] In some alternative implementations, to avoid overly extreme probability estimates (e.g., 0.0) when the sample size is small (e.g., W is small or some intervals have 0 samples), smoothing techniques can be used, such as adding a small pseudo-count (e.g., 1) to the frequency N_k of each interval, i.e., Laplace smoothing.

[0081] Step 403: Based on the probability distribution of the safety margin interval, calculate the risk entropy value of a single project using the information entropy formula, and collect them into a time series set of risk entropy for the single project.

[0082] Specifically, after obtaining the probability distribution p_e_k(t) of the safety margin interval at time step t, the definition of information entropy (Shannon entropy) is applied to quantify the uncertainty of this distribution. The formula for calculating the single-project risk entropy H_single_e(t) is: H_single_e(t) = -Σ_k(p_e_k(t) * log(p_e_k(t)));

[0083] Here, Σ_k represents the summation over all safety margin intervals k=1,2,3,4... The logarithm can be base 2 or base e; in this invention, a consistent base is sufficient. To prevent the logarithmic calculation from being meaningless when the probability p_e_k(t) is 0, it is agreed that when p_e_k(t)=0, p_e_k(t)*log(p_e_k(t))=0.

[0084] The higher the value of the risk entropy H_single_e(t), the greater the uncertainty of the safety margin (i.e., the margin value is dispersed across multiple intervals), which in itself represents a risk. Conversely, the lower the value, the more certain the safety margin falls within a single interval.

[0085] Continuing with the example from step 402, whose probability distribution is [0.1, 0.5, 0.4, 0.0], the single-project risk entropy (base 2) at that moment is calculated as follows:

[0086] H_single_e(t)=-(0.1*log2(0.1)+0.5*log2(0.5)+0.4*log2(0.4)+0*log2(0));

[0087] H_single_e(t)=-(0.1*(-3.32)+0.5*(-1.0)+0.4*(-1.32));

[0088] H_single_e(t)=-(-0.332-0.5-0.528)=1.36bits;

[0089] By repeating this calculation for each project e at all time steps t, the time series set of risk entropy for the single project can be obtained.

[0090] Step 404: Simultaneously, the safety margin interval corresponding to the safety margin value of each time step is marked as the single-project safety margin interval label set.

[0091] Specifically, this step is a parallel marking process. In step 401, the instantaneous value of M_safe_e(t) is discretized. This step is to save the result of this discretization.

[0092] For example, at time step t, the M_safe_e(t) of a certain project is 0.01875. According to the rule in step 401, this value falls into interval 2 (the margin critical interval). Therefore, the interval label of the project at that time is interval 2 (or its numerical index 2).

[0093] By aggregating the interval labels of all projects across all time steps, the single-project safety margin interval label set is formed. This label set will be used in subsequent embodiments to construct discrete risk states and build model input features.

[0094] Example 5 describes how to combine the single-project risk entropy time series set generated in Example 4 with the pre-constructed graph structure data to construct the subsystem risk entropy time series set.

[0095] Step 501: Extract engineering topology relationships and hydraulic connections from the pre-constructed entropy flux Markov graph structure data.

[0096] Specifically, the entropy flux Markov graph structured data is a pre-established data structure describing the watershed risk propagation network. In this step, it is necessary to query and extract the static relationships between projects from this structured data.

[0097] Engineering topology refers to the spatial layout and connection of engineering projects in a watershed network. For example, project A is located upstream of project B, projects C and D operate in series, and projects E and F are connected in parallel for flood discharge.

[0098] Hydraulic connections represent the mutual influence of topological relationships on hydrodynamics, such as the degree of impact of upstream reservoir discharge on the water level of downstream dikes, and the propagation time lag of the confluence of two sub-basins.

[0099] This step also includes an implicit process: dividing all engineering nodes within the basin into several subsystems based on the aforementioned topological relationships and hydraulic connections. A subsystem is a collection of projects with close hydraulic connections or that share a common flood control task. For example, a group of reservoirs on the same main stream can be one subsystem, while a group of dikes and pumping stations surrounding the same protected area (such as a city) can be another subsystem.

[0100] Step 502: Based on the engineering topology relationship, determine the topology weight of each project within the subsystem.

[0101] Specifically, after assigning a project to a subsystem s, it is necessary to determine the topology weight w_topo_s_e corresponding to each project e within that subsystem. This weight reflects the importance or risk contribution of project e within subsystem s.

[0102] The determination of this topological weight can be based on a variety of factors. For example:

[0103] Based on project location: Projects located at key nodes of the subsystem (such as confluence points or downstream outlets) may have a higher weight w_topo_s_e.

[0104] Based on project scale: projects with larger flood control capacity C_design_e and wider protection range may have a higher weight w_topo_s_e.

[0105] Based on the consequences of failure: If the failure has a more severe impact on downstream or protected areas, the weight w_topo_s_e may be higher.

[0106] All weights are normalized within the subsystem, i.e., Σ_ew_topo_s_e=1.

[0107] Step 503: Based on the hydraulic connection, determine the coupling strength coefficient between the engineering components within the subsystem.

[0108] Specifically, this step is used to quantify the mutual influence strength between engineering pairs (e.g., e1 and e2) within a subsystem s that have direct hydraulic or scheduling coupling relationships. This strength is defined as the coupling strength coefficient c_s_e1_e2.

[0109] For example, if the flood discharge operation of project e1 (upstream reservoir) immediately and significantly affects the water level of project e2 (downstream levee), then the coupling strength coefficient c_s_e1_e2 between them is high. If the two projects e3 and e4 are located on different tributaries of the same subsystem and have little mutual influence, then c_s_e3_e4 is low or 0. This coefficient can be calibrated based on hydrodynamic model simulations or historical data statistics.

[0110] Step 504: The time series set of risk entropy of the single project is weighted according to the topological weight, and the coupling strength coefficient between the projects and the difference in risk entropy of the single project are combined to jointly construct the time series set of risk entropy of the subsystem.

[0111] Specifically, this step is the core calculation process of the subsystem risk entropy H_sub_s(t). In this embodiment, a calculation formula that integrates weighted averaging and coupling differences is preferably used:

[0112] H_sub_s(t)=Σ_e(w_topo_s_e*H_single_e(t))+λ_s*Σ_(e1,e2)(c_s_e1_e2*|H_single_e1(t)-H_single_e2(t)|);

[0113] Where H_sub_s(t) is the risk entropy of subsystem s at time step t.

[0114] The first term Σ_e(w_topo_s_e*H_single_e(t)) is the topology-weighted average of the single-project risk entropy (from Example 4) of all projects within the subsystem, reflecting the overall average risk level of the subsystem.

[0115] The second term is the coupling difference term. Σ_(e1,e2) represents the summation of all coupled pairs (e1,e2) within subsystem s. |H_single_e1(t)-H_single_e2(t)| is the absolute difference in risk entropy between projects e1 and e2. c_s_e1_e2 is the coupling strength coefficient between them. λ_s is the coupling strength adjustment coefficient of subsystem s (a hyperparameter used to balance the dimensions and importance of the first and second terms).

[0116] This formula not only considers the average risk of the subsystem (first term) but also introduces risk imbalance as an independent source of risk (second term). For example, if within a subsystem, the entropy of an engineering process e1 is high (e.g., 0.9, close to failure), while the entropy of its closely coupled engineering process e2 is low (e.g., 0.1, very safe), this huge entropy difference |H_single_e1(t)-H_single_e2(t)|=0.8 itself represents the instability of the system and the potential risk of cascading failure. This formula quantifies this risk by incorporating it into the total entropy of the subsystem through the second term, thus providing a more comprehensive measure of risk.

[0117] As an example, assume that subsystem s contains only two projects, e1 and e2. Its topological weights are w_topo_s_1 = 0.6 and w_topo_s_2 = 0.4. The coupling coefficient between them is c_s_e1_e2 = 0.8. The adjustment coefficient is λ_s = 0.1.

[0118] At time step t, H_single_e1(t) = 1.36, H_single_e2(t) = 0.5.

[0119] Then H_sub_s(t) = (0.6*1.36 + 0.4*0.5) + 0.1*(0.8*|1.36-0.5|);

[0120] H_sub_s(t)=(0.816+0.2)+0.1*(0.8*0.86);

[0121] H_sub_s(t)=1.016+0.1*(0.688)=1.016+0.0688=1.0848;

[0122] By repeating this calculation for all subsystems s and all time steps t, the time series set of risk entropy for the subsystems can be obtained.

[0123] In some alternative implementations, the subsystem risk entropy can also be calculated in other forms. For example, a simplified approach is to use only the first term, i.e., H_sub_s(t) = Σ_e(w_topo_s_e * H_single_e(t)). Another alternative is to use a nonlinear function instead of linear weighting, such as H_sub_s(t) = max_e(H_single_e(t)), where the risk of the subsystem is determined by the project with the highest risk within it.

[0124] Example 6 describes how to construct a system risk entropy time series, and how to integrate all intermediate results generated during the calculation process of this invention into the real-time safety margin state set and the real-time hierarchical risk entropy time series front end required for subsequent steps.

[0125] Step 601: Merge the subsystem risk entropy time series set with the engineering flood control design parameter set to construct the system risk entropy time series.

[0126] Specifically, this step involves further converging the subsystem risk entropy H_sub_s(t) upwards to construct the overall system risk entropy H_sys(t) at the watershed level.

[0127] This construction process requires incorporating the exposure and vulnerability parameter set and the engineering flood control design parameter set extracted in Example 1 to determine the weight of each subsystem s. In this example, it is preferable to assign two weights to each subsystem s:

[0128] Exposure weight w_exp_s: This weight is determined based on the exposure and vulnerability parameter set and is used to quantify the severity of the consequences that may result from the failure of subsystem s. For example, a subsystem s1 that protects an important city and a large population would have a high exposure weight w_exp_s1 (e.g., 0.5); while another subsystem s2 that protects farmland would have a lower exposure weight w_exp_s2 (e.g., 0.1).

[0129] Control weight w_ctrl_s: This weight is determined based on the set of flood control design parameters of the project and is used to quantify the flood control capacity and control importance of subsystem s. For example, it can be determined comprehensively based on factors such as the scheduling flexibility, engineering redundancy, and design flood control level within the subsystem. A subsystem with strong scheduling capacity, high redundancy, and high flood control level may also have its control weight w_ctrl_s set accordingly (e.g., set to 0.8).

[0130] The formula for calculating the system risk entropy H_sys(t) can be: H_sys(t)=Σ_s(w_exp_s*w_ctrl_s*H_sub_s(t)).

[0131] Here, Σ_s represents the summation over all subsystems s within the watershed. The formula means that the total risk entropy of the watershed is a weighted sum of the risk entropies H_sub_s(t) of all subsystems, where the weighting considers both the consequences of subsystem failure (exposure weight) and its own importance (prevention weight). A subsystem with high H_sub_s(t), high exposure weight w_exp_s, and high prevention weight w_ctrl_s will be the main contributor to the total risk entropy H_sys(t).

[0132] By repeating this calculation at all time steps t, the time series of the system risk entropy can be obtained.

[0133] In some alternative implementations, the system risk entropy H_sys(t) can also be expressed in an unweighted manner, for example, H_sys(t) = max_s(H_sub_s(t)), that is, the total system risk is determined by the subsystem with the highest risk.

[0134] Step 602: Combine the single-project safety margin time series set with the single-project safety margin interval label set to form the real-time safety margin status set; and gather the single-project risk entropy time series set, the subsystem risk entropy time series set, and the system risk entropy time series to form the front end of the real-time hierarchical risk entropy time series.

[0135] Specifically, this step involves standardizing and integrating all the results generated in Examples 3 to 6 to create a unified data structure that can be input into subsequent prediction models (such as Example 8).

[0136] The real-time safety margin state set is a data structure. Within this structure, for each project e and each time step t, two key pieces of information are stored:

[0137] The original safety margin value M_safe_e(t) from Example 3 (e.g., 0.01875);

[0138] Discrete interval labels from Example 4 (e.g., interval 2).

[0139] This combination allows the model to obtain both accurate continuous values ​​and discrete features with clear risk level implications.

[0140] The aforementioned real-time hierarchical risk entropy time series front section is another data structure. "Hierarchical" refers to its inclusion of the three risk levels constructed in this invention; "front section" refers to its inclusion of all data from the historical starting point to the current time step t, used as the input sequence for predicting the future. This data structure aggregates: all single-project risk entropies H_single_e(t); all subsystem risk entropies H_sub_s(t); and system risk entropies H_sys(t).

[0141] In summary, Examples 3, 4, 5, and 6 fully illustrate the entire process of calculating the real-time safety margin state set and the first segment of the real-time hierarchical risk entropy time series from the raw data.

[0142] Example 7 details the specific method for constructing the pre-built entropy flux Markov graph structure data referenced in Example 2. This data structure is pre-built based on historical data and watershed physical characteristics before model training and real-time operation.

[0143] Step 701: Abstract the watershed topology and establish a graph structure. Before constructing the entropy flux Markov graph structure data, it is first necessary to abstract the physical structure of the watershed. Specifically, the basic geographic information data and full-element information tensor of the watershed obtained in Example 1 are read, and the watershed is abstracted into a watershed engineering topology graph structure data composed of engineering nodes and river segment edges based on the river connectivity, upstream and downstream water level control relationships, and engineering scheduling paths.

[0144] The engineering node can represent a single project (e.g., a reservoir dam, a key levee section, or a group of floodgates) or a local confluence area. The river edge represents the hydraulic transmission path or scheduling control path between nodes. Preferably, each edge is assigned attributes describing its physical characteristics, such as average flow direction, average propagation time delay (e.g., 30 minutes), and typical flow range.

[0145] Step 702: Obtain engineering nodes from the engineering flood control design parameter set, and establish an entropy budget equation set for any engineering node. This step is the theoretical basis for constructing Markov graphs.

[0146] Specifically, the present invention regards the single-project risk entropy H_single_j(t) calculated in Example 4 as a quantity that can flow and dissipate between nodes j in the topology graph, thereby establishing the transmission relationship of risk entropy between nodes.

[0147] It should be noted that the analogy of risk entropy (an information theory concept) to a fluid physical quantity is an analogical modeling method constructed to address the spatiotemporal propagation of risk in flood control scheduling. It is not equivalent to energy conservation in thermodynamics or mass conservation in fluid mechanics. Under certain assumptions (e.g., the transmission relationship of risk between adjacent projects can be quantified by an entropy flux function), this model can serve as an effective approximation describing the spatiotemporal evolution of risk.

[0148] Based on this, for any engineering node j, its entropy budget equation can be constructed as follows:

[0149] H_single_j(t+1)=H_single_j(t)+Σ_iF_entropy_i_j(t)-Σ_kF_entropy_j_k(t)-D_entropy_j(t);

[0150] Where H_single_j(t+1) and H_single_j(t) are the single-engineering risk entropies of node j at adjacent time steps (t+1 and t), respectively (from the historical sequence of Example 4). Σ_iF_entropy_i_j(t) represents the aggregated entropy flux inflow from all upstream nodes i to node j. Σ_kF_entropy_j_k(t) represents the aggregated entropy flux outflow from node j to all downstream nodes k. D_entropy_j(t) represents the local entropy dissipation term at node j at time step t.

[0151] The local entropy dissipation term D_entropy_j(t) has a clear physical meaning; it represents the reduction (i.e., entropy reduction) in local risk uncertainty caused by proactive flood control measures (such as temporary reinforcement or activating pumping stations for forced drainage) or scheduling operations (such as activating flood diversion channels). This term can be fitted as a function of risk entropy level and scheduling operation instructions using historical data, forming an entropy dissipation parameter set.

[0152] Step 703, calibrate the entropy flux coefficient. To make the above entropy budget equations (established for all nodes j) solvable, it is necessary to first calibrate the aggregated entropy flux inflow F_entropy_i_j(t).

[0153] In this embodiment, the entropy flux from node i to node j is defined as being related not only to the risk entropy H_single_i(t) of upstream node i, but also to the hydraulic connections (such as flow rate) and scheduling operations (such as gates) between them. A preferred formula for calculating the entropy flux is:

[0154] F_entropy_i_j(t)=α_ij*H_single_i(t)*g_flow(Q_ij(t),U_ij(t))

[0155] Here, α_ij is the entropy transfer coefficient from node i to node j, reflecting the coupling strength of hydraulic or scheduling operations. This coefficient is a core parameter that needs to be calibrated using historical data. H_single_i(t) is the single-project risk entropy of upstream node i. g_flow(...) is a normalization function used to map the actual flow Q_ij(t) on edge i->j and the associated scheduling operation intensity U_ij(t) (e.g., gate opening) to their contribution to entropy transfer.

[0156] By reading the hierarchical risk entropy time series and the full-element information tensor from historical flood cases, sample periods containing typical flood processes can be selected. Using the relationship between entropy and flow changes in these sample periods, the entropy transfer coefficients α_ij on all edges are fitted and estimated using the least squares method or other optimization algorithms, ultimately forming an entropy flux coefficient matrix set.

[0157] Specifically, the calibration process of the entropy transfer coefficient α_ij can be carried out using the following method:

[0158] At least 30 complete flood events from historical flood cases were selected as training samples. For each sample, the single-project risk entropy change ΔH_single_j(t), the upstream node risk entropy H_single_i(t), the actual flow rate Q_ij(t) on the edge, and the scheduling operation intensity U_ij(t) were extracted for adjacent time steps.

[0159] Construct an optimization objective function, for example: minimize Σ_samples Σ_t Σ_j (ΔH_single_j(t)- Σ_i(α_ij * H_single_i(t) * g_flow(Q_ij(t), U_ij(t))) + D_entropy_j(t))^2;

[0160] The optimization problem is solved using the least squares method or gradient descent algorithm to obtain the entropy transfer coefficient α_ij that minimizes the error between the predicted entropy change and the actual entropy change. In a preferred embodiment, the entropy transfer coefficient α_ij typically ranges from 0 to 1, with larger values ​​(e.g., 0.6 to 0.9) for sides with tight hydraulic coupling between upstream and downstream, and smaller values ​​(e.g., 0.1 to 0.3) for sides with weaker hydraulic coupling.

[0161] Step 704: Calculate the entropy increment and derive the entropy flux Markov transition matrix.

[0162] After establishing the entropy budget equation (step 702) and calibrating the entropy flux coefficient (step 703), the theoretical entropy increment ΔH_single_j(t) = H_single_j(t+1) - H_single_j(t) at any time step t can be calculated using historical data.

[0163] The entropy increment ΔH_single_j(t) and the current safety margin state (M_safe_j(t) from Example 6) jointly drive the evolution of the risk state of the engineering node. This step models this evolution process as a Markov process.

[0164] First, a finite number of discrete risk states need to be defined. These states can be defined based on the safety margin interval labels in Example 4. For example, sufficient margin, low margin, critical margin, and negative margin can be defined as four discrete states S1, S2, S3, and S4.

[0165] The entropy flux Markov transition matrix P_j has elements P_j(m,n) representing the probability that node j transitions from the current state Sm to the next state Sn. This probability P_j(m,n) is derived based on the entropy increment ΔH_single_j(t) and the safety margin M_safe_j(t), rather than a simple historical frequency count.

[0166] For example, a probability function P_j(m,n)=f_markov(ΔH_single_j(t),M_safe_j(t)) can be constructed. Using historical data, the parameters of this function f_markov can be solved using maximum likelihood estimation or Bayesian estimation methods. For instance, when the entropy increment ΔH_single_j(t) is significantly positive (entropy increases sharply), the probability P_j(2,3) derived from the model of transitioning from low margin (S2) to critical margin (S3) will be significantly increased.

[0167] By performing this derivation on all nodes j, the set of entropy flux Markov transition matrices can be obtained.

[0168] Step 705: Integrate graph structure data. Finally, the data generated in the above steps are uniformly encoded and integrated. The watershed engineering topology graph structure data (nodes and edges) generated in step 701, the entropy flux coefficient matrix set calibrated in step 703, the entropy dissipation parameter set determined in step 702, and the entropy flux Markov transition matrix set derived in step 704 are uniformly encoded and stored.

[0169] The resulting data packet is the pre-constructed entropy flux Markov graph structure data. This data packet will be used as input to a prediction model in subsequent embodiments to express the propagation pattern of entropy on the topological network and to deduce failure chains.

[0170] Example 8: This example describes the construction and training method of the entropy-aware dual-stream prediction model used for the training completion of step 202 in Example 2.

[0171] Step 801: Construct an entropy-aware dual-stream network structure. This network structure is the core of the prediction model of this invention. Preferably, the structure includes two parallel but interactive data streams: a physical process stream and a risk structure stream.

[0172] The purpose of the described physical process flow is to extract the dynamic characteristics of flood evolution and engineering response. Specifically, this flow takes the dynamic time series in the full-element information tensor generated in Example 1 as input, such as rainfall, inflow, water level, reservoir capacity, gate opening, etc. The network structure of this flow can preferably adopt a deep learning network that is good at processing time series, such as a multi-layer stack of Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), or Temporal Convolutional Network (TCN).

[0173] The risk structure flow aims to express the propagation pattern of entropy on the topological network. Specifically, the flow takes as input the hierarchical risk entropy time series set and safety margin state set generated in Example 6, and the entropy flux Markov graph structure data generated in Example 7. Since the input naturally has a graph structure (engineering nodes and river segment edges), the network structure of the flow can preferably adopt a graph neural network (GNN), such as a graph attention network (GAT) or a graph convolutional network (GCN). These networks can capture the spatial dependence and propagation characteristics of risk entropy on the watershed topology.

[0174] Step 802: An explicit entropy channel is established between the physical process flow and the risk structure flow to achieve controlled two-flow information interaction. The detailed implementation of this step is further described in Embodiment 9. Its purpose is to ensure that the evolution of the physical process (such as a rainstorm) can be perceived by the risk structure, while changes in the risk state (such as a surge in entropy in a project) can, in turn, constrain the prediction of the physical process.

[0175] Step 803: The entropy-aware two-stream network structure is trained using a joint loss function to obtain the trained entropy-aware two-stream prediction model. The detailed implementation of this step is further explained in Example 10.

[0176] During the training phase, it is first necessary to construct training and validation datasets. Specifically, multiple complete flood events are selected from historical data and processed according to the methods in Examples 1 to 7 to generate corresponding input features (full-element information tensors, the first segment of the hierarchical risk entropy time series, etc.) and supervision labels (true values ​​of hierarchical risk entropy at future time steps, entropy budget equation constraints, historical failure chain samples, etc.). These data pairs are then divided into training and validation sets according to time.

[0177] Then, configure the optimization algorithm (such as the Adam optimizer), and set the initial learning rate, batch size, and upper limit of training epochs.

[0178] In a preferred embodiment, the initial learning rate can be set in the range of 0.0001 to 0.001, for example, 0.0005; the batch size can be set in the range of 16 to 128, for example, 32 or 64; and the upper limit of the training epochs can be set in the range of 100 to 500 epochs. For the Adam optimizer, β1 can use a default value of 0.9, and β2 can use a default value of 0.999. If an early stopping strategy is adopted, training can be terminated and the current optimal model parameters saved when the validation loss no longer decreases within 10 to 20 consecutive epochs.

[0179] In one alternative implementation, for temporal networks (such as LSTM or GRU) in physical process flows, the hidden layer dimension can be set in the range of 64 to 256, and the number of layers can be set in the range of 2 to 4; for graph neural networks (such as GAT or GCN) in risk structure flows, the hidden dimension can be set in the range of 32 to 128, and the number of attention heads can be set in the range of 4 to 8.

[0180] In each training round, the training dataset is batch-wise input into the entropy-aware two-stream network for forward propagation, and the joint loss is calculated. Then, backpropagation is performed based on the joint loss to update the network parameters. Simultaneously, the validation loss is periodically evaluated on the validation dataset to monitor for overfitting. When the validation loss no longer decreases significantly or an early stopping strategy is triggered, the current network parameters are saved, resulting in the trained entropy-aware two-stream prediction model.

[0181] Example 9: This example, as a preferred implementation of step 802 in Example 8, elaborates in detail how to set an explicit entropy channel to achieve controlled two-stream information interaction.

[0182] Step 901: Construct an entropy channel feature vector for each node, which includes risk entropy, safety margin, and entropy flux indicators.

[0183] Specifically, at each time step t during model training and inference, for each engineering node e in the topology graph, a set of key risk indicators is extracted from the data in Examples 6 and 7. This feature vector may exemplarily include:

[0184] The current single-project risk entropy is H_single_e(t);

[0185] The current single-project safety margin is M_safe_e(t).

[0186] The current subsystem risk entropy H_sub_s(t) (where e belongs to s);

[0187] The current system risk entropy H_sys(t);

[0188] The entropy increment ΔH_single_e(t) within the recent historical window;

[0189] Entropy flux inflow / outflow within the recent historical window ΣF_in, ΣF_out.

[0190] These indicators are concatenated to form a high-dimensional entropy channel feature vector V_entropy_e(t). This vector comprehensively summarizes the full-level risk state of node e at time t.

[0191] Step 902: The entropy channel feature vector is used as an independent entropy feature label and embedded into the physical process flow and the risk structure flow.

[0192] Specifically, this step involves introducing V_entropy_e(t) as a special token into the two-stream network. In deep learning implementations, this can be viewed as additionally concatenating the V_entropy_e(t) vector into the input sequence of the physical process stream (such as LSTM) and the node features of the risk structure stream (such as GNN).

[0193] In this way, when processing the feature of a water level of 78.5 meters, the physical process flow can simultaneously see an entropy feature marker, which tells it that the current risk entropy is 1.36, indicating a critical state. Correspondingly, when processing the feature of a risk entropy of 1.36, the risk structure flow can also see the information about the water level of 78.5 meters from the physical process flow.

[0194] Step 903, and by configuring a restricted attention mechanism in the entropy-aware two-stream network structure, the interaction of the two-stream information is constrained by the entropy feature label.

[0195] Specifically, simply performing feature concatenation (as in step 902) constitutes a weak interaction. To achieve controlled, strong interaction, this embodiment preferably employs a restricted attention mechanism.

[0196] In modern deep learning networks (such as Transformer or graph attention networks), information interaction is achieved through attention heads. In this embodiment, constraints are imposed on some attention heads in the network, causing them to compute attention weights only between entropy feature labels and other node features.

[0197] For example, in a physical process flow, an entropy-guided attention head is configured that, when calculating future water levels, must forcefully (or with high weight) pay attention to the entropy feature tag V_entropy_e(t).

[0198] In the risk structure flow, a physically aware attention head is configured that, when calculating the future risk entropy, must forcibly focus on features from the physical process flow (such as predicted future flows).

[0199] Through this explicit and constrained attention design, the present invention constructs a dedicated entropy channel. This channel ensures that the prediction of physical flood evolution is always constrained by the current and future risk entropy states; at the same time, the prediction of risk entropy propagation can always perceive changes in the actual hydrological process, thereby achieving deep and controlled fusion of the two flows.

[0200] Example 10 describes a preferred implementation of step 803 in Example 8, and elaborates on the specific structure and implementation of the joint loss function. This joint loss function is the optimization objective for training the entropy-aware dual-stream prediction model.

[0201] In this embodiment, the joint loss function L_total is preferably a weighted combination of three loss terms with different physical or statistical meanings. L_total = β1 * L_entropy + β2 * L_flux + β3 * L_chain, where β1, β2, and β3 are weighting coefficients used to balance the influence of each loss term.

[0202] The hierarchical risk entropy prediction error loss term L_entropy. This loss term is a major component of the joint loss function and is used to quantify the deviation between the model's output of the predicted set of future hierarchical risk entropy and the historical hierarchical risk entropy time series (i.e., the true labels).

[0203] Specifically, the loss term L_entropy is preferably calculated using Mean Squared Error (MSE) or its variants (such as smoothed L1 loss). An example of its plain text calculation formula is as follows:

[0204] L_entropy=Σ_t,e(H_single_e_pred(t)-H_single_e_true(t))^2+Σ_t,s(H_sub_s_pred(t)-H_sub_s_true(t))^2+Σ_t(H_sys_pred(t)-H_sys_true(t))^2;

[0205] Where H_single_e_pred(t) is the single-project risk entropy predicted by the model at a future time step t, and H_single_e_true(t) is the corresponding true single-project risk entropy in the training dataset (calculated from the results of Example 6). H_sub_s and H_sys correspond to the predicted and true values ​​of the subsystem and the system, respectively. Σ_t,e,Σ_t,s,Σ_t represent the summation over all prediction time steps and over all projects, subsystems, and systems, respectively.

[0206] The purpose of the L_entropy loss term is to drive the model parameters to be updated in a direction that can more accurately regress and predict the true values ​​of the risk entropy at the three levels of the future, through backpropagation.

[0207] The entropy balance constraint loss term L_flux is used to penalize deviations between the future hierarchical risk entropy prediction result set and the future entropy flux prediction result set from the pre-established entropy balance equation set (from Example 7).

[0208] Specifically, the loss term is a physics-informed constraint based on a physical analogy model. It first obtains the model's single-engineering risk entropy predictions H_single_j_pred(t+1) and H_single_j_pred(t) for adjacent future time steps, as well as the entropy flux inflow predictions Σ_iF_entropy_i_j_pred(t) and outflow predictions Σ_kF_entropy_j_k_pred(t) for the same time step, and combines them with the known local entropy dissipation term D_entropy_j(t) (from Example 7).

[0209] Calculate the residual Residual_j(t) of the entropy budget equation:

[0210] Residual_j(t)=H_single_j_pred(t+1)-H_single_j_pred(t)-(Σ_iF_entropy_i_j_pred(t)-Σ_kF_entropy_j_k_pred(t)-D_entropy_j(t));

[0211] Ideally, the model's predictions should result in zero residuals. Therefore, the L_flux loss term is the sum of squared residuals at all nodes j and time step t: L_flux = Σ_t,j(Residual_j(t))^2;

[0212] The purpose of the L_flux loss term is to ensure that while the model learns to predict entropy and entropy flux, it must also learn to abide by the conservation relationship between them as they are propagated on the topological network. This avoids the model producing prediction results that are significantly contrary to the laws of risk propagation, thereby improving the model's generalization ability and physical consistency.

[0213] The failure chain consistency loss term L_chain is used to calibrate the consistency between the set of future failure chain probability distributions and the failure paths (or historical disaster samples) inferred based on the pre-built entropy flux Markov graph structure data.

[0214] Specifically, the loss term L_chain is preferably calculated using cross-entropy loss or Kullback-Leibler divergence. One output header of the model (as described in Example 8) outputs a probability distribution vector P_pred(t), representing the probability of each predefined failure chain (e.g., chain 1: failure of dam A, chain 2: failure of dam A leading to the collapse of dam B) occurring at a future time step t.

[0215] At the same time, a true probability distribution P_true(t) is obtained from the training data (such as historical real disaster samples or steady-state probabilities derived from Markov graphs).

[0216] L_chain is the cross-entropy between P_pred(t) and P_true(t):

[0217] L_chain=-Σ_t,c(P_true_c(t)*log(P_pred_c(t))); where c represents traversing all predefined failure chains.

[0218] The purpose of the L_chain loss term is to ensure that the model can not only predict the total amount of risk (L_entropy), but also accurately predict the pattern of risk (i.e., the most likely way in which cascading failures will occur), making the prediction results more meaningful for decision-making.

[0219] During model training, the optimal combination of weight coefficients β1, β2, and β3 is obtained by fine-tuning them on the validation set (e.g., through grid search or Bayesian optimization). For example, a possible setting is β1=0.6, β2=0.2, and β3=0.2, indicating that the prediction accuracy of risk entropy is the primary factor, supplemented by physical constraints and failure chain constraints.

[0220] Example 11: This example describes how, during the real-time operation phase, the model trained in Example 10 is used to extrapolate real-time data to obtain the future risk prediction set.

[0221] Step 1101: Real-time acquisition of full-element information tensors and engineering parameters. During the real-time flood control scheduling operation phase, the method of this invention will continue to run. The system will read the latest monitoring data streams in real time from the various monitoring systems described in Example 1 (such as hydrological stations, rain gauges, and engineering automation systems), and acquire the latest meteorological forecast data.

[0222] The system will employ the same data processing flow as in Example 1, including quality control, spatiotemporal benchmark unification, assimilation and fusion, and attribute mapping, to process this new data and combine it with near-historical data to form a real-time, all-element information tensor that updates over time. Simultaneously, it will load the statically stored set of engineering flood control design parameters.

[0223] Step 1102: Real-time safety margin and real-time hierarchical risk entropy calculation. This step prepares the current state input for the prediction model. The system will obtain the real-time full-element information tensor and the engineering flood control design parameter set from step 1101.

[0224] It is important to emphasize that this step is a calculation, not a prediction. It quickly calculates the current and recent historical timeframes within a short time window (e.g., based on data from the past hour).

[0225] Real-time single-project safety margin time series; real-time single-project risk entropy time series; real-time subsystem risk entropy time series; real-time system risk entropy time series.

[0226] These calculation results are integrated into the real-time safety margin state set and the real-time hierarchical risk entropy time series front segment defined in Example 6.

[0227] Step 1103, entropy-aware two-stream model forward inference. This is the core of this embodiment, namely, the real-time prediction of the execution model.

[0228] Load three key data points:

[0229] The entropy-aware dual-stream prediction model trained in Example 10 (including all its network parameters).

[0230] The entropy flux Markov graph structure data (including topology and transition matrix) pre-constructed in Example 7.

[0231] The real-time input data generated in step 1102 (real-time safety margin state set and real-time hierarchical risk entropy time series front segment).

[0232] The system packages real-time input data into the input format required by the model according to the entropy channel construction method defined in Example 9.

[0233] Perform one forward pass. That is, feed the real-time input data into the trained prediction model. Based on its learned parameters, the model directly calculates the output result without backpropagation or parameter updates. This calculation process outputs the results of the three prediction heads defined in Example 8:

[0234] Future-level risk entropy prediction result set (e.g., H_single_e, H_sub_s, H_sys predicted values ​​every 15 minutes within the next 6 hours).

[0235] Future entropy flux prediction results set (e.g., F_entropy_i_j prediction values ​​on the critical path in the next 6 hours).

[0236] The probability distribution set of future failed chains (e.g., the probability of P(chain1), P(chain2), etc. in the next 6 hours).

[0237] The system aligns and calibrates these three sets of prediction results over time (e.g., associates them with specific prediction timestamps) and encapsulates them into a unified data structure, namely the future risk prediction set. This prediction set will serve as direct input for downstream early warning and decision-making steps.

[0238] In some alternative implementations, to assess the reliability of the prediction results, an evaluation of prediction uncertainty can be added at this step. For example, Monte Carlo Dropout can be used to run forward propagation multiple times (e.g., 30 times) during the inference phase while retaining the Dropout, thereby obtaining a distribution of a series of prediction results. Based on this, a confidence interval (e.g., a 95% confidence interval) for each prediction value (e.g., H_sub_s_pred(t+1)) can be calculated, and this uncertainty evaluation result can be output together.

[0239] If the real-time input data is found to be missing or abnormal, such as the data interruption of a monitoring station exceeding the preset time, such as 30 minutes, the system may adopt one of the following strategies: (1) fill with the most recent valid data; (2) use data from adjacent monitoring stations for interpolation estimation; (3) mark "data incomplete, prediction results for reference only" in the warning results.

[0240] If a computational anomaly occurs during the model inference process, such as numerical overflow or model output exceeding a reasonable range, the system can trigger the backup plan: (1) roll back to the valid prediction result of the previous moment; (2) switch to a simplified rule-based early warning mechanism; (3) send a system anomaly alarm to the dispatcher to prompt manual intervention.

[0241] Example 12: This example describes how to combine the future risk prediction set generated in Example 11 with a preset threshold to generate a dynamic risk warning result set.

[0242] Step 1201: Extract future risk prediction data. The system first reads the future risk prediction set generated in Example 11. From this dataset, two key pieces of information required for this step are extracted:

[0243] Future-level risk entropy prediction result set (e.g., the H_single_e_pred(t) and H_sub_s_pred(t) sequences of all projects e and subsystems s within the next 6 hours);

[0244] The probability distribution set of future failure chains (e.g., the sequence of P_pred_c(t) within the next 6 hours).

[0245] The future entropy flux prediction result set is mainly reserved for Implementation Example 13 for scheme evaluation.

[0246] Step 1202: Perform preliminary risk level determination based on entropy level threshold. The system compares the future level risk entropy prediction result set with the preset entropy level threshold.

[0247] The entropy level thresholds are pre-determined based on historical extreme events and watershed flood control safety management requirements. These thresholds map continuous risk entropy values ​​to discrete, easily understandable warning levels for dispatchers. For example, a set of thresholds for the risk entropy H_sub_s of a subsystem can be defined as follows:

[0248] Level 4 (Reliable Zone): H_sub_s < 0.5;

[0249] Level 3 (Guard Zone): 0.5 <= H_sub_s < 1.0;

[0250] Level 2 (Crisis Zone): 1.0 <= H_sub_s < 1.5;

[0251] Level 1 (Critical Section): H_sub_s>=1.5;

[0252] The system will compare the predicted risk entropy H_pred(t) of each engineering node e and subsystem s with the above threshold at each future time step t (e.g., t+1h, t+2h...).

[0253] For example, if the model predicts that the risk entropy of a subsystem s1 after 2 hours is H_sub_s1_pred(t+2h)=1.15, then the initial risk level of the subsystem at that moment is classified as level 2 (crisis zone).

[0254] Step 1203 involves correcting and confirming the risk level by incorporating the failure chain probability. This step is crucial to the invention, as it avoids potential misjudgments or insufficient information arising from relying solely on entropy thresholds. The system will use the future failure chain probability distribution set extracted in step 1201 to correct or confirm the preliminary judgment result from step 1202.

[0255] Specifically, the combination methods include, but are not limited to:

[0256] Risk Confirmation and Escalation: Assume that step 1202 identifies subsystem s1 as a crisis zone. Simultaneously, the system discovers in the failure chain probability set that the predicted probability P_pred_c(t+2h) of the critical failure chain related to s1 (e.g., failure of dam A within s1) is also significantly increased (e.g., >30%). This strongly confirms the crisis zone identification, and the warning result will be marked as a crisis zone (high risk: dam A failure chain).

[0257] Risk downgrading or identification: Suppose step 1202 determines that subsystem s2 is a crisis zone, but the failure chain probability set shows that the probabilities of all known critical failure chains related to s2 are low (e.g., <5%). This may indicate that although the system uncertainty (entropy) is high, no clear, high-probability failure mode has formed. The warning result may be marked as a crisis zone, i.e., high uncertainty, but no clear failure chain, prompting dispatchers to conduct a comprehensive investigation rather than targeting a specific path.

[0258] Risk Coverage and Supplementation: Assume that the discrimination subsystem s3 in step 1202 is only in the alert zone (with low entropy), but the failure chain probability set shows that the probability P_pred_c(t+2h) of a black swan failure chain (e.g., the cascading failure of reservoirs B and C) has increased abnormally and slightly (e.g., from 0.1% to 2%). The system can specifically address this situation by adding a special attention marker to the alert zone's warning: an abnormal probability marker for the BC cascading failure chain.

[0259] Through the above combined judgment, the system finally forms the dynamic risk warning result set. This result set is a multi-dimensional information table that not only indicates the warning level (such as crisis zone) for each future time step and each project node, but may also include the specific failure mode that leads to that level (such as the failure chain of Dam A), thus providing dispatchers with richer and more interpretable warning information.

[0260] Example 13: This example describes how to generate an emergency dispatch scheme set based on early warning results and prediction sets, with specific constraints and objectives.

[0261] Step 1301: In response to the dynamic risk warning results, simulate candidate emergency dispatch operations.

[0262] Specifically, the triggering condition for this step is that the dynamic risk warning result set generated in Example 12 indicates an increase in the risk level, for example, that a certain subsystem or component will enter the alert zone, crisis zone or critical zone within a future time step (e.g., 3 hours later).

[0263] Once triggered, the system will read three key input data: 1) the dynamic risk warning result set generated in Example 12 (used to determine the severity, location, and possible failure modes of the risk); 2) the future risk prediction set generated in Example 11 (in particular, the prediction sequence of future entropy flux and future hierarchical risk entropy); 3) the engineering flood control design parameter set constructed in Example 1 (used to clarify the physical constraints of the operation).

[0264] Based on the above inputs, the system begins to simulate candidate emergency dispatch operations. These operations must first meet the hard constraints of the engineering flood control design parameter set. For example, the pre-release operation of a reservoir cannot exceed the safe discharge capacity of the downstream river channel or its own structural limitations; the opening speed of the gates must be within their physical limits; and the dispatch instructions must comply with the basic provisions of the basin flood control plan.

[0265] The candidate emergency dispatch operations can be a single operation or a sequence. For example: within 0-2 hours, the discharge flow of Reservoir A increases to 500 cubic meters per second; or Option B: within 1-3 hours, Gate B opens for flood diversion, while Pumping Station C operates at full capacity. These candidate options can be obtained by matching from a predefined dispatch rule base, or by obtaining them from a historical successful dispatch case base through case-based reasoning, or by initially generating them through optimization algorithms (such as genetic algorithms or particle swarm optimization).

[0266] Step 1302: Evaluate candidate emergency dispatch operations to select a set of emergency dispatch schemes.

[0267] Specifically, this step involves evaluating, ranking, and filtering the multiple (e.g., N) candidate scheduling operation schemes generated in step 1301. The core basis for the evaluation is two objectives:

[0268] Meeting safety margin constraints: This is a bottom-line or hard constraint objective. That is, the execution of a scheduling plan (such as the pre-release of water from Reservoir A) cannot come at the expense of the safety of another project (such as Dike C). The system will evaluate whether, after executing the plan, the safety margin M_safe_e(t) (defined and calculated in Example 3) of all relevant projects within the basin remains above a preset safety lower limit (e.g., M_safe_e(t) > 0.01) within the predicted future timeframe. Any plan that causes the safety margin of any project to exceed the lower limit will be directly eliminated.

[0269] Entropy reduction objective: This is an optimization or soft-constraint objective. That is, while satisfying all safety margin constraints, the scheduling scheme should most effectively reduce the overall future risk uncertainty of the system. This entropy reduction can be quantified by multiple metrics, such as minimizing the peak value of the system risk entropy H_sys(t), or minimizing the integral value of the system risk entropy H_sys(t) over the entire warning period.

[0270] To achieve the above assessment, the present invention preferably utilizes the future risk prediction set in Example 11 and the prediction model trained in Example 10. The system will simulate the execution of each candidate scheme (e.g., scheme i).

[0271] A preferred simulation method is to use scheme i (e.g., the new discharge process of reservoir A at times t+1, t+2...) as a new set of operational features, update the real-time full-element information tensor in embodiment 11, and then re-invoke the entropy-aware dual-flow prediction model for forward inference. The model will output a new set of future hierarchical risk entropy prediction results and future entropy flux prediction results under the intervention of scheme i.

[0272] The system then analyzes this new set of predictions. First, it checks whether the safety margins of all projects (which can be indirectly reflected by the new entropy predictions or calculated through the model) meet the constraints (Objective 1). Then, it calculates the total entropy reduction (Objective 2) resulting from this scheme i compared to taking no action (i.e., the original prediction).

[0273] For example, scheme i reduces the peak value of the total system entropy H_sys(t) by 20% over the next 6 hours, and all engineering safety margins meet the constraints. Scheme j, on the other hand, only reduces the entropy by 5%, and causes the safety margin of levee D to fall below the lower limit.

[0274] In this way, the system sorts the N candidate solutions, eliminates solutions that do not meet the safety margin (such as solution j), and selects one or more (k) solutions with the best entropy reduction effect from the solutions that meet the conditions (such as solution i).

[0275] These k preferred solutions are organized into the emergency dispatch scheme set. This scheme set can be further visualized, for example, in the form of dispatch instruction sheets, risk entropy evolution comparison curves under different schemes, and provided to the flood control dispatch personnel in the basin as a scientific decision-making basis for consultation, confirmation and final execution, thus completing a complete closed loop from data to early warning to decision-making.

[0276] Suppose a watershed contains 3 reservoirs (A1, A2, A3), 2 sections of dikes (B1, B2) and 1 flood diversion gate (C1). A1 is located at the upstream end, A2 and A3 are connected in parallel in the middle reaches, B1 and B2 are connected in series in the downstream, and C1 is used for flood diversion to protect downstream cities.

[0277] At 2:00 PM on a certain day, heavy rainfall occurred upstream, with a forecast of 120 mm of cumulative rainfall in the next 6 hours.

[0278] Step 1401: Data Acquisition and Construction of the Full-Element Tensor. The system acquires real-time data from hydrological monitoring stations: the current water level of Reservoir A1 is 78.5m (design flood control level 80.0m), with an inflow of 800 m³ / s; the water levels of Reservoirs A2 and A3 are 65.2m and 66.8m, respectively; the monitoring water level of Dike B1 is 12.3m (design allowable water level 15.0m). The system constructs a full-element information tensor for the current moment.

[0279] Step 1402, Safety Margin and Risk Entropy Calculation. Following the method in Example 3, calculate the safety margin for each project:

[0280] Reservoir A1: M_safe_A1 = (80.0-78.5) / 80.0 = 0.01875 (margin critical interval);

[0281] B1 embankment: M_safe_B1 = (15.0-12.3) / 15.0 = 0.18 (sufficient margin range).

[0282] Following the method in Example 4, the single-project risk entropy is calculated based on the safety margin sample distribution over the past 60 minutes:

[0283] H_single_A1(t) = 1.36 bits (high, risky);

[0284] H_single_B1(t) = 0.32 bits (low, safe for now).

[0285] Step 1403, Model Inference. Input the current state into the entropy-aware dual-stream prediction model, and the model outputs the prediction results for the next 6 hours:

[0286] The risk entropy of reservoir A1 is predicted to reach 1.85 bits (entering the crisis zone) at t+3h.

[0287] The probability of the "A1 failure → B1 collapse" failure chain is predicted to be 35%.

[0288] Step 1404, Warning Generation. Based on a preset threshold (crisis zone ≥ 1.0 bits), the system generates a warning:

[0289] "Warning: Reservoir A1 is expected to enter the crisis zone in 3 hours, with a 35% probability of a chain failure. Immediate emergency measures are recommended."

[0290] Step 1405, Emergency Plan Generation. System Evaluation Candidate Solutions:

[0291] Option 1: Reservoir A1 immediately pre-releases water to 500 m³ / s → Predicted entropy decreases to 1.2 bits, but B1 margin decreases to 0.12 (still meets the constraint).

[0292] Option 2: Open the C1 flood diversion gate → the prediction entropy decreases to 1.0 bits, and the B1 margin remains at 0.15.

[0293] The system recommends option 2 because it has a better entropy reduction effect and less impact on downstream processes.

[0294] In summary, this invention introduces physical safety margin as an anchor point for risk quantification. By first calculating the relative margin between engineering design capability and real-time load, and then calculating the information entropy of the probability distribution of this safety margin, a three-tiered risk entropy system is constructed, from single project to subsystem to system. This approach directly links risk entropy to the physical safety reserve of the project, solving the problems of traditional failure probability measurements being vague and unintuitive, and giving risk assessment a solid physical basis.

[0295] Furthermore, an entropy flux Markov graph model was constructed. By establishing an analogous entropy budget equation, risk entropy is regarded as a flux that can flow and dissipate in the watershed topology network, and the Markov transition matrix is ​​derived based on this entropy budget relationship. This successfully transforms isolated node risks into a spatiotemporally coupled dynamic propagation process, realizing the quantitative prediction of the evolution path of cascading failure risks.

[0296] An entropy-aware dual-flow prediction model was designed. This model processes flood and entropy information in parallel through physical process flow and risk structure flow. It enforces a deep interaction and constraint between physical predictions (e.g., flow rate) and risk states (e.g., entropy increase) through explicit entropy channels and a restricted attention mechanism. Simultaneously, a joint loss function incorporating entropy conservation ensures that the model's predictions simultaneously conform to physical laws and risk propagation laws, achieving a strong physical-risk coupling prediction effect.

Claims

1. A method for real-time flood control scheduling risk dynamic early warning and emergency scheme making of a river basin, characterized in that, The method comprises the following steps: Obtaining real-time full-element information tensor and engineering flood control design parameter set; Accordingly, the real-time safety margin state set and the real-time hierarchical risk entropy time sequence segment are calculated; The real-time full-element information tensor, the real-time safety margin state set, and the pre-constructed entropy flux Markov graph structure data are input into the trained entropy perception double-flow prediction model for reasoning to obtain a future risk prediction set including a future hierarchical risk entropy prediction result set, a future entropy flux prediction result set, and a future failure chain probability distribution set; Combined with the real-time hierarchical risk entropy time sequence segment, the future risk prediction set, and the preset entropy level threshold, a dynamic risk early warning result set is generated; Based on the dynamic risk early warning result set, the future risk prediction set, and the engineering flood control design parameter set, an emergency scheduling scheme set is generated to meet the safety margin constraint and the entropy reduction target.

2. The method of claim 1, wherein, Accordingly, the real-time safety margin state set and the real-time hierarchical risk entropy time sequence segment are calculated, including: The single-project safety margin time sequence set is calculated based on the real-time full-element information tensor and the engineering flood control design parameter set; Based on the single-project safety margin time sequence set, the single-project risk entropy time sequence set is generated by estimating the safety margin interval probability distribution, and the single-project safety margin interval label set is obtained; The subsystem risk entropy time sequence set is constructed based on the single-project risk entropy time sequence set and the pre-constructed entropy flux Markov graph structure data; The system risk entropy time sequence is constructed by fusing the subsystem risk entropy time sequence set and the engineering flood control design parameter set; The real-time safety margin state set is formed by combining the single-project safety margin time sequence set and the single-project safety margin interval label set; and the real-time hierarchical risk entropy time sequence segment is formed by collecting the single-project risk entropy time sequence set, the subsystem risk entropy time sequence set, and the system risk entropy time sequence.

3. The method of claim 2, wherein, The single-project safety margin time sequence set is calculated based on the real-time full-element information tensor and the engineering flood control design parameter set, including: For each engineering node, the engineering design flood control capacity is extracted from the engineering flood control design parameter set; The real-time actual flood load of the corresponding engineering node is extracted from the real-time full-element information tensor; Based on the engineering design flood control capacity and the real-time actual flood load, the safety margin value is calculated and collected as the single-project safety margin time sequence set.

4. The method of claim 2, wherein, Based on the single-project safety margin time sequence set, the single-project risk entropy time sequence set is generated by estimating the safety margin interval probability distribution, and the single-project safety margin interval label set is obtained, including: According to the preset safety margin interval division rule, the safety margin values in the single-project safety margin time sequence set are discretized; At each time step, the conditional probability of the safety margin value falling into each safety margin interval is calculated to obtain the safety margin interval probability distribution; Based on the safety margin interval probability distribution, the single-project risk entropy value is calculated by applying the information entropy formula and is collected as the single-project risk entropy time sequence set; At the same time, the safety margin interval corresponding to the safety margin value at each time step is marked as the single-project safety margin interval label set.

5. The method of claim 2, wherein, The single-engineering risk entropy time series set is combined with the pre-constructed entropy flux Markov graph structure data to construct a subsystem risk entropy time series set, including: From the pre-constructed entropy flux Markov graph structure data, the engineering topological relationship and hydraulic connection are extracted; Based on the engineering topological relationship, the topological weight of each engineering in the subsystem is determined; Based on the hydraulic connection, the coupling strength coefficient between the engineering in the subsystem is determined; The single-engineering risk entropy time series set is weighted according to the topological weight, and the coupling strength coefficient between the engineering and the single-engineering risk entropy difference are fused to jointly construct the subsystem risk entropy time series set.

6. The method of claim 1, wherein, The pre-constructed entropy flux Markov graph structure data is obtained by the method comprising the following steps: For the engineering nodes in the basin, an entropy balance equation set is established, which quantifies the relationship between the time variation of single-engineering risk entropy, upstream entropy flux inflow, downstream entropy flux outflow and local entropy dissipation; Based on the entropy balance equation set, the entropy increment of each engineering node is calculated; And according to the entropy increment and the corresponding safety margin state, the entropy flux Markov transition matrix for describing the evolution of the node risk state is derived.

7. The method of claim 6, wherein, Establishing the entropy balance equation set includes: For any engineering node, the single-engineering risk entropy at the subsequent time step is constructed; the single-engineering risk entropy at the subsequent time step is determined by the single-engineering risk entropy at the current time step, the aggregated entropy flux inflow from all upstream nodes, the aggregated entropy flux outflow to all downstream nodes, and the local entropy dissipation term.

8. The method of claim 1, wherein, The trained entropy-aware double-flow prediction model is obtained by the training method comprising the following steps: An entropy-aware double-flow network structure is constructed, including a physical process flow for extracting flood evolution features and a risk structure flow for expressing entropy topological propagation patterns; An explicit entropy channel is set between the physical process flow and the risk structure flow to realize controlled double-flow information interaction; A joint loss function is used to train the entropy-aware double-flow network structure to obtain the trained entropy-aware double-flow prediction model.

9. The method of claim 8, wherein, Setting an explicit entropy channel to realize controlled double-flow information interaction includes: An entropy channel feature vector containing risk entropy, safety margin and entropy flux indicators is constructed for each node; The entropy channel feature vector is embedded as an independent entropy feature label in the physical process flow and the risk structure flow; And by configuring a restricted attention mechanism in the entropy-aware double-flow network structure, the double-flow information interaction is constrained by the entropy feature label.

10. The method of claim 8, wherein, The joint loss function includes: A hierarchical risk entropy prediction error loss term for quantifying the deviation between the future hierarchical risk entropy prediction result set output by the model and the historical hierarchical risk entropy time series; An entropy balance conservation constraint loss term for penalizing the deviation of the future hierarchical risk entropy prediction result set and the future entropy flux prediction result set from the pre-established entropy balance equation set; A failure chain consistency loss term for calibrating the consistency between the future failure chain probability distribution set and the failure path deduced based on the pre-constructed entropy flux Markov graph structure data.

Images